Large language models (LLMs) are increasingly utilized in clinical reasoning and risk assessment. However, their interpretive reliability in critical and indeterminate domains such as psychiatry remains unclear. Prior work has identified algorithmic biases and prompt sensitivity in these systems, raising concerns about how contextual information may influence model outputs, but there remains no systematic way to assess these, especially in the psychiatric domain. We propose an approach for reliability auditing downstream LLM tasks by structuring evaluation around the impact of prompt design and the inclusion of medically insignificant inputs on predicted hospitalization risk scores, which is often the first downstream AI clinical-decision-making task. In our audit, a cohort of synthetic patient profiles (n = 50) is generated, each consisting of 15 clinically relevant features and up to 50 clinically insignificant features, across four prompt reframings (neutral, logical, human impact, clinical judgment). We audit four LLMs (Gemini 2.5 Flash, LLaMa 3.3 70b, Claude Sonnet 4.6, GPT-4o mini), and our results show that including medically insignificant variables resulted in a statistica
Large language models (LLMs) are increasingly used for decision support in healthcare, but clinical evidence is often incomplete or evolving. When the available information is insufficient to support a reliable answer, models should request clarification or abstain rather than provide unsupported responses. Existing medical benchmarks, however, typically assume that complete information is available upfront. We introduce Safe-Psych, a sequential benchmark for evaluating how LLMs handle evolving diagnostic uncertainty in clinical psychiatry. Safe-Psych contains over 1,000 real-world psychiatric clinical notes segmented to simulate incremental evidence disclosure, with psychiatrist-derived action labels at each stage: DIAGNOSE, CLARIFY, or ABSTAIN. We evaluate multiple state-of-the-art LLMs in full-information and sequential settings. Our findings show that capability does not ensure calibration: even strong models struggle under incomplete clinical information, with under-abstention exceeding 60% for most models and safety-aware prompting reducing premature commitment only by shifting errors toward excessive abstention. In sequential evaluation, models frequently diagnose before suf
Recent reports indicate that sustained interaction with conversational artificial intelligence (AI) systems can, in a small subset of users, contribute to the emergence or stabilisation of delusional experience. Existing accounts typically attribute such cases either to individual vulnerability or to failures of safety engineering. These explanations are incomplete. Drawing on phenomenology, psychiatry, and cognitive neuroscience, this paper argues that the risk arises from the relational and ontological structure of the interaction itself. Conversational AI generates ontological dissonance: a conflict between the appearance of relational presence and the absence of any subject capable of sustaining it. Maintained through a communicative double bind and amplified by attentional asymmetries, this dissonance tends, under conditions of affective vulnerability, to stabilise into a technologically mediated analogue of folie a deux. This account explains why explicit disclaimers often fail to disrupt delusional involvement and clarifies the ethical and clinical implications for the design and use of conversational AI.
Modelling psychological disorders in artificial agents offers both a testbed for computational psychiatry and a lens on the failure modes of affective control. Prior work induces one or two disorders in a reinforcement learning (RL) agent by hand-tuned reward shaping, labels the behaviour post hoc, and reports single runs. We recast disorder modelling as dose-controllable manipulation of cognitive appraisal signals in an appraisal-guided PPO agent, expressing seven disorders (anxiety, mania, obsessive-compulsive checking, depression, impulsivity, addiction, and post-traumatic stress) each as a single knob grounded in a computational psychiatry account, with each symptom measured by a preregistered assay mapped to a recognised paradigm. Across more than a thousand runs (10 seeds, four controls, 95% confidence intervals) every disorder shows a graded, monotone dose-response that no control reproduces. Beyond these induced effects, three findings emerge that were not written into the reward: the disorders self-organise into a two-dimensional affective space in which mania mirrors anxiety; removing a knob remits reward distortion disorders (mania, checking, addiction) but not avoidance
Large language models (LLMs) hold transformative potential for medical decision support yet their application in psychiatry remains constrained by hallucinations and superficial reasoning. This limitation is particularly acute in light-parameter LLMs which are essential for privacy-preserving and efficient clinical deployment. Existing training paradigms prioritize linguistic fluency over structured clinical logic and result in a fundamental misalignment with professional diagnostic cognition. Here we introduce ClinMPO, a reinforcement learning framework designed to align the internal reasoning of LLMs with professional psychiatric practice. The framework employs a specialized reward model trained independently on a dataset derived from 4,474 psychiatry journal articles and structured according to evidence-based medicine principles. We evaluated ClinMPO on a unseen subset of the benchmark designed to isolate reasoning capabilities from rote memorization. This test set comprises items where leading large-parameter LLMs consistently fail. We compared the ClinMPO-aligned light LLM performance against a cohort of 300 medical students. The ClinMPO-tuned Qwen3-8B model achieved a diagnos
Computational psychiatry faces a fundamental trade-off: traditional reinforcement learning (RL) models offer interpretability but lack behavioral realism, while large language model (LLM) agents generate realistic behaviors but lack structural interpretability. We introduce BioLLMAgent, a novel hybrid framework that combines validated cognitive models with the generative capabilities of LLMs. The framework comprises three core components: (i) an Internal RL Engine for experience-driven value learning; (ii) an External LLM Shell for high-level cognitive strategies and therapeutic interventions; and (iii) a Decision Fusion Mechanism for integrating components via weighted utility. Comprehensive experiments on the Iowa Gambling Task (IGT) across six clinical and healthy datasets demonstrate that BioLLMAgent accurately reproduces human behavioral patterns while maintaining excellent parameter identifiability (correlations $>0.67$). Furthermore, the framework successfully simulates cognitive behavioral therapy (CBT) principles and reveals, through multi-agent dynamics, that community-wide educational interventions may outperform individual treatments. Validated across reward-punishme
Large language models (LLMs) offer significant potential in enhancing psychiatric practice, from improving diagnostic accuracy to streamlining clinical documentation and therapeutic support. However, existing evaluation resources heavily rely on small clinical interview corpora, social media posts, or synthetic dialogues, which limits their clinical validity and fails to capture the full complexity of diagnostic reasoning. In this work, we introduce PsychiatryBench, a rigorously curated benchmark grounded exclusively in authoritative, expert-validated psychiatric textbooks and casebooks. PsychiatryBench comprises eleven distinct question-answering tasks ranging from diagnostic reasoning and treatment planning to longitudinal follow-up, management planning, clinical approach, sequential case analysis, and multiple-choice/extended matching formats totaling 5,188 expert-annotated items. {\color{red}We evaluate a diverse set of frontier LLMs (including Google Gemini, DeepSeek, Sonnet 4.5, and GPT 5) alongside leading open-source medical models such as MedGemma using both conventional metrics and an "LLM-as-judge" similarity scoring framework. Our results reveal substantial gaps in clin
Large language models (LLMs) are increasingly proposed as scalable solutions to the global mental health crisis. But their deployment in psychiatric contexts raises a distinctive ethical concern: the problem of atypicality. Because LLMs generate outputs based on population-level statistical regularities, their responses -- while typically appropriate for general users -- may be dangerously inappropriate when interpreted by psychiatric patients, who often exhibit atypical cognitive or interpretive patterns. We argue that standard mitigation strategies, such as prompt engineering or fine-tuning, are insufficient to resolve this structural risk. Instead, we propose dynamic contextual certification (DCC): a staged, reversible and context-sensitive framework for deploying LLMs in psychiatry, inspired by clinical translation and dynamic safety models from artificial intelligence governance. DCC reframes chatbot deployment as an ongoing epistemic and ethical process that prioritises interpretive safety over static performance benchmarks. Atypicality, we argue, cannot be eliminated -- but it can, and must, be proactively managed.
Mobile technology (e.g., mobile phones and wearable devices) provides scalable methods for collecting physiological and behavioral biomarkers in patients' naturalistic settings, as well as opportunities for therapeutic advancements and scientific discoveries regarding the etiology of psychiatric illness. Continuous data collection through mobile devices generates highly complex data: entangled multivariate time series of outcomes, exposures, and covariates. Missing data is a pervasive problem in biomedical and social science research, and Ecological Momentary Assessment (EMA) data in psychiatric research is no exception. However, the complex data structure of multivariate time series and their non-stationary nature make missing data a major challenge for proper inference. Additional historical information included in time series analyses exacerbates the issue of missing data and also introduces problems for confounding adjustment. The majority of existing imputation methods are either designed for stationary time series or for longitudinal data with limited follow-up periods. The limited work on non-stationary time series either focuses on missing exogenous information or ignores t
The emergency department (ED) is a high stress environment with increased risk of clinician bias exposure. In the United States, Black patients are more likely than other racial/ethnic groups to obtain their first schizophrenia (SCZ) diagnosis in the ED, a highly stigmatizing disorder. Therefore, understanding the link between clinician bias exposure and psychiatric outcomes is critical for promoting nondiscriminatory decision-making in the ED. This study examines the association between clinician bias exposure and psychiatric diagnosis using a sample of patients with anxiety, bipolar, depression, trauma, and SCZ diagnoses (N=29,005) from a diverse, large medical center. Clinician bias exposure was quantified as the ratio of negative to total number of sentences in psychiatric notes, labeled using a large language model (Mistral). We utilized logistic regression to predict SCZ diagnosis when controlling for patient demographics, risk factors, and negative sentence ratio (NSR). A high NSR significantly increased one's odds of obtaining a SCZ diagnosis and attenuated the effects of patient race. Black male patients with high NSR had the highest odds of being diagnosed with SCZ. Our f
The application of AI in psychiatric diagnosis faces significant challenges, including the subjective nature of mental health assessments, symptom overlap across disorders, and privacy constraints limiting data availability. To address these issues, we present MoodAngels, the first specialized multi-agent framework for mood disorder diagnosis. Our approach combines granular-scale analysis of clinical assessments with a structured verification process, enabling more accurate interpretation of complex psychiatric data. Complementing this framework, we introduce MoodSyn, an open-source dataset of 1,173 synthetic psychiatric cases that preserves clinical validity while ensuring patient privacy. Experimental results demonstrate that MoodAngels outperforms conventional methods, with our baseline agent achieving 12.3% higher accuracy than GPT-4o on real-world cases, and our full multi-agent system delivering further improvements. Evaluation in the MoodSyn dataset demonstrates exceptional fidelity, accurately reproducing both the core statistical patterns and complex relationships present in the original data while maintaining strong utility for machine learning applications. Together, the
Studying psychiatric illness has often been limited by difficulties in connecting symptoms and behavior to neurobiology. Computational psychiatry approaches promise to bridge this gap by providing formal accounts of the latent information processing changes that underlie the development and maintenance of psychiatric phenomena. Models based on these theories generate individual-level parameter estimates which can then be tested for relationships to neurobiology. In this review, we explore computational modelling approaches to one key aspect of health and illness: affect. We discuss strengths and limitations of key approaches to modelling affect, with a focus on reinforcement learning, active inference, the hierarchical gaussian filter, and drift-diffusion models. We find that, in this literature, affect is an important source of modulation in decision making, and has a bidirectional influence on how individuals infer both internal and external states. Highlighting the potential role of affect in information processing changes underlying symptom development, we extend an existing model of psychosis, where affective changes are influenced by increasing cortical noise and consequent i
Forensic mental health care involves the treatment of individuals with severe mental disorders who have committed violent offences. These settings are often characterized by high levels of bureaucracy, risk avoidance, and restricted autonomy. Patients frequently experience a profound loss of control over their lives, leading to heightened psychological stress-sometimes resulting in isolation as a safety measure. In this study, we explore how co-design can be used to collaboratively develop a companion robot that helps monitor and regulate stress while maintaining tracking of the patients' interaction behaviours for long-term intervention. We conducted four co-design workshops in a forensic psychiatric clinic with patients, caregivers, and therapists. Our process began with the presentation of an initial speculative prototype to therapists, enabling reflection on shared concerns, ethical risks, and desirable features. This was followed by a creative ideation session with patients, a third workshop focused on defining desired functions and emotional responses, and we are planning a final prototype demo to gather direct patient feedback. Our findings emphasize the importance of empowe
Perceptual multistability, observed across species and sensory modalities, offers valuable insights into numerous cognitive functions and dysfunctions. For instance, differences in temporal dynamics and information integration during percept formation often distinguish clinical from non-clinical populations. Computational psychiatry can elucidate these variations, through two primary approaches: (i) Bayesian modeling, which treats perception as an unconscious inference, and (ii) an active, information-seeking perspective (e.g., reinforcement learning) framing perceptual switches as internal actions. Our synthesis aims to leverage multistability to bridge these computational psychiatry subfields, linking human and animal studies as well as connecting behavior to underlying neural mechanisms. Perceptual multistability emerges as a promising non-invasive tool for clinical applications, facilitating translational research and enhancing our mechanistic understanding of cognitive processes and their impairments.
The classification of clinical notes into specific diagnostic categories is critical in healthcare, especially for mental health conditions like Anxiety and Adjustment Disorder. In this study, we compare the performance of various Artificial Intelligence models, including both traditional Machine Learning approaches (Random Forest, Support Vector Machine, K-nearest neighbors, Decision Tree, and eXtreme Gradient Boost) and Deep Learning models (DistilBERT and SciBERT), to classify clinical notes into these two diagnoses. Additionally, we implemented three oversampling strategies: No Oversampling, Random Oversampling, and Synthetic Minority Oversampling Technique (SMOTE), to assess their impact on model performance. Hyperparameter tuning was also applied to optimize model accuracy. Our results indicate that oversampling techniques had minimal impact on model performance overall. The only exception was SMOTE, which showed a positive effect specifically with BERT-based models. However, hyperparameter optimization significantly improved accuracy across the models, enhancing their ability to generalize and perform on the dataset. The Decision Tree and eXtreme Gradient Boost models achiev
Mental disorders have become a significant global public health issue, while the shortage of psychiatrists and inefficient training systems severely hinder the accessibility of mental health services. This paper designs and implements an artificial intelligence-based training system for psychiatrists. By integrating technologies such as large language models, knowledge graphs, and expert systems, the system constructs an intelligent and standardized training platform. It includes six functional modules: case generation, consultation dialogue, examination prescription, diagnostic decision-making, integrated traditional Chinese and Western medicine prescription, and expert evaluation, providing comprehensive support from clinical skill training to professional level assessment.The system adopts a B/S architecture, developed using the Vue.js and Node.js technology stack, and innovatively applies deep learning algorithms for case generation and doctor-patient dialogue. In a clinical trial involving 60 psychiatrists at different levels, the system demonstrated excellent performance and training outcomes: system stability reached 99.95%, AI dialogue accuracy achieved 96.5%, diagnostic ac
Background: Symptom rating scales in psychiatry are limited by reliance on self-report, and lack of predictive power. Actigraphy, a passive wearable-based method for measuring sleep and physical activity, offers objective, high-resolution behavioral data that may better reflect symptom fluctuations, but most studies have focused on narrow diagnostic groups or fixed time windows, limiting clinical translation. Objective: To examine whether actigraphy-derived sleep and activity features correlate with psychiatric symptom severity in a transdiagnostic psychiatric sample, and to identify which features are most clinically relevant across multiple temporal resolutions. Methods: We present a feasibility case series study with preliminary data from eight outpatients enrolled in the DeeP-DD study, a transdiagnostic study of digital phenotyping. Participants wore GENEActiv actigraphy devices and symptom severity was measured using a variety of validated scales. We performed intra-individual Spearman correlations and inter-individual repeated measures correlations across daily, weekly, monthly, and full-duration averages. Results: Intra-individual analyses revealed that later rise times were
Social agents with finitely nested opponent models are vulnerable to manipulation by agents with deeper recursive capabilities. This imbalance, rooted in logic and the theory of recursive modelling frameworks, cannot be solved directly. We propose a computational framework called $\aleph$-IPOMDP, which augments the Bayesian inference of model-based RL agents with an anomaly detection algorithm and an out-of-belief policy. Our mechanism allows agents to realize that they are being deceived, even if they cannot understand how, and to deter opponents via a credible threat. We test this framework in both a mixed-motive and a zero-sum game. Our results demonstrate the $\aleph$-mechanism's effectiveness, leading to more equitable outcomes and less exploitation by more sophisticated agents. We discuss implications for AI safety, cybersecurity, cognitive science, and psychiatry.
Flexible machine learning tools are increasingly used to estimate heterogeneous treatment effects. This paper gives an accessible tutorial demonstrating the use of the causal forest algorithm, available in the R package grf. We start with a brief non-technical overview of treatment effect estimation methods, focusing on estimation in observational studies; the same techniques can also be applied in experimental studies. We then discuss the logic of estimating heterogeneous effects using the extension of the random forest algorithm implemented in grf. Finally, we illustrate causal forest by conducting a secondary analysis on the extent to which individual differences in resilience to high combat stress can be measured among US Army soldiers deploying to Afghanistan based on information about these soldiers available prior to deployment. We illustrate simple and interpretable exercises for model selection and evaluation, including targeting operator characteristics curves, Qini curves, area-under-the-curve summaries, and best linear projections. A replication script with simulated data is available at https://github.com/grf-labs/grf/tree/master/experiments/ijmpr
Integrating cognitive ergonomics with LLMs is crucial for improving safety, reliability, and user satisfaction in human-AI interactions. Current LLM designs often lack this integration, resulting in systems that may not fully align with human cognitive capabilities and limitations. This oversight exacerbates biases in LLM outputs and leads to suboptimal user experiences due to inconsistent application of user-centered design principles. Researchers are increasingly leveraging NLP, particularly LLMs, to model and understand human behavior across social sciences, psychology, psychiatry, health, and neuroscience. Our position paper explores the need to integrate cognitive ergonomics into LLM design, providing a comprehensive framework and practical guidelines for ethical development. By addressing these challenges, we aim to advance safer, more reliable, and ethically sound human-AI interactions.